Full-band vibration and noise suppression method for vehicle active suspension

Abstract

The invention relates to a full-band vibration and noise suppression method for a vehicle active suspension. The method comprises the steps that a high-frequency quarter vehicle active suspension model used for describing the dynamic relation between suspension vibration and tire acoustic response is constructed; based on the high-frequency quarter vehicle active suspension model, a model prediction control strategy is designed, and model prediction control force meeting system constraint conditions is obtained through solving; constructing a reinforcement learning controller, and introducing the model prediction control force as expert reference information into a reinforcement learning training process to obtain an active suspension control model; and inputting the current system state into the active suspension control model, and outputting to obtain an active suspension control quantity which is used for correspondingly controlling the working state of the active suspension. Compared with the prior art, the low-frequency comfort and the medium-high frequency noise suppression performance can be considered, collaborative optimization of vehicle vibration and in-vehicle noise is achieved, and the comprehensive sound vibration comfort of the whole vehicle is improved.

Description

Technical Field

[0001] This invention relates to the field of vehicle active suspension control technology, and in particular to a method for suppressing vibration and noise across the entire frequency band of a vehicle active suspension. Background Technology

[0002] With the increasing demands for improved vibration and acoustic comfort in automobiles, vehicle vibration and noise issues have gradually become important factors affecting overall vehicle quality. Road excitation is transmitted to the vehicle body structure through the tires and suspension system, easily causing body vibration and in-vehicle road noise, which adversely affects the driving and riding experience.

[0003] Vehicle vibration and noise have different mechanisms of action across different frequency ranges: low-frequency vibrations primarily affect ride comfort and vehicle handling stability, while mid- to high-frequency vibrations are converted into in-vehicle noise through structural coupling. Therefore, from a system perspective, achieving coordinated control of vehicle vibration and noise across the entire frequency band has become an important direction for the development of vehicle chassis technology.

[0004] Currently, electronically controlled suspension systems provide an effective means of vehicle vibration control by actively adjusting suspension forces. However, existing suspension control research is mostly based on traditional vehicle dynamics models, focusing on the suppression effect of body vibration in the low-frequency range. However, traditional vehicle dynamics models typically do not adequately consider the high-frequency structural characteristics and acoustic effects of tires, making it difficult to accurately describe mid-to-high frequency vibrations and their transmission to in-vehicle noise. Furthermore, with the application of intelligent control methods in suspension systems, some control strategies, in the absence of high-frequency dynamic constraints, may introduce high-frequency changing control forces, thereby exciting mid-to-high frequency vibrations in the suspension and tire systems, further exacerbating in-vehicle noise problems. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art by providing a method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension, which can balance low-frequency comfort with mid-to-high-frequency noise suppression performance, achieve synergistic optimization of vehicle vibration and in-vehicle noise, and improve the overall acoustic and vibration comfort of the vehicle.

[0006] The objective of this invention can be achieved through the following technical solution: a method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension, comprising the following steps: S1. Construct a high-frequency quarter-vehicle active suspension model to describe the dynamic relationship between suspension vibration and tire acoustic response. S2. Based on the high-frequency quarter-vehicle active suspension model, design a model predictive control strategy and solve for the model predictive control force that satisfies the system constraints. S3. Construct a reinforcement learning controller and introduce the model-predicted control force as expert reference information into the reinforcement learning training process to obtain an active suspension control model. S4. Input the current system state into the active suspension control model and output the active suspension control quantity, which is used to control the working state of the active suspension accordingly.

[0007] Furthermore, step S1 specifically involves constructing a high-frequency quarter-vehicle active suspension model based on suspension structure dynamics, tire rigid ring characteristics, and tire acoustic cavity modal coupling effect.

[0008] Further, step S1 includes the following process: S11. Introduce the tire rigid ring characteristics into the suspension structure dynamics model to describe the structural dynamic behavior of the tire under medium and high frequency operating conditions, and establish a high frequency structural dynamics model of the tire-wheel system. S12. Considering the influence of the structural deformation of the tire under static load on the acoustic cavity mode, the main acoustic modes of the tire acoustic cavity are equivalently modeled as simple harmonic dynamic elements, and the original degenerate mode is decomposed into vertical mode and lateral mode. The acoustic mode that plays a dominant role in the acoustic response under vehicle operation is selected as the modeling object. The dynamic behavior of the tire acoustic cavity is described in the coordinate system that rotates with the tire to establish the tire acoustic cavity model. S13. Based on the tire rigid ring model and the tire acoustic cavity model, establish the mechanical coupling relationship between the acoustic cavity vibration and the tire-wheel system, and introduce the equivalent force generated by the acoustic cavity vibration into the tire structure dynamics model to characterize the bidirectional coupling effect between the tire acoustic cavity mode and the suspension structure. S14. Integrate the tire structure dynamics model, tire acoustic cavity model, suspension system and vehicle body structure dynamics model to construct a high-frequency quarter vehicle active suspension model.

[0009] Furthermore, the high-frequency structural dynamics model of the tire-wheel system in step S11 is specifically as follows: in, z s , z u and z b These represent the displacements of the sprung mass, the unsprung mass, and the rigid ring, respectively. m s Represents the sprung mass. m a Represents the unsprung mass of the section without a rigid ring. m b Represents the mass of the rigid ring. kb and c b These represent the wheel-ring stiffness and damping, respectively, connecting the rim and the rigid ring. k ct This represents the grounding stiffness connecting the rigid ring and the ground. k s ,and c s These represent suspension stiffness and damping, respectively.

[0010] Furthermore, the tire acoustic cavity model in step S12 is specifically as follows: in, X and Z This indicates the motion of the center of mass of the enclosed air within the acoustic cavity of a tire; ω 0 represents the natural frequency of the tire's acoustic cavity in the unloaded state; ζ The damping ratio; f 0 represents the static load applied along the Z-axis; a and b Describe the offset of the natural frequencies of the vertical and lateral modes under load, respectively, where a < 0, b > 0; c This is the proportionality coefficient. f r This represents the random dynamic component in the tire-road interaction force.

[0011] Furthermore, the equivalent force generated by the vibration of the acoustic cavity in step S13 is specifically as follows: in, w e Indicates the effective width of the wheel; r w Indicates the wheel radius; ρ 0 represents the density of the gas inside the tire in equilibrium.

[0012] Furthermore, the high-frequency quarter-vehicle active suspension model in step S14 is specifically as follows: in, F a This is for active suspension control.

[0013] Further, step S2 includes the following process: S21. Based on the high-frequency quarter-vehicle active suspension model, obtain system state information to construct a predictive control optimization model containing system constraints. S22. Solve the predictive control optimization model within the finite prediction time domain to obtain the model predictive control force, which is used as the reference control input for the active suspension.

[0014] Furthermore, the system status information includes vehicle body acceleration, suspension travel, and tire dynamic load, specifically the vehicle body vibration response. Body abruptness Vehicle speed The relative displacement of the suspension z0 -z u and its changing characteristics - .

[0015] Furthermore, step S3 includes the following process: S31. Construct the state observations of the reinforcement learning controller as follows: s t =[ , z 0 -z u , - The state observations include system state information that characterizes the low-frequency comfort and mid-to-high-frequency vibration and noise characteristics of the vehicle. S32. Define the action space of the reinforcement learning controller as the active suspension control force. F a Its value range is limited by the physical constraints of the suspension structure and actuators. F a ∈[ F min , F max ]; S33. Construct a composite reward function that includes time-domain performance metrics and frequency-domain performance metrics. R ( s t , a t )= R time ( s t , a t )+ R freq ( s t , a t (Among them, time-domain performance indicators) R time ( s t ,a t This is used to constrain the low-frequency ride comfort and suspension operating conditions of a vehicle; frequency domain performance indicators. R freq ( s t , a t (This is used to characterize the vibration response characteristics related to in-vehicle noise within a target frequency band, which includes the 0-20Hz, 20-50Hz and 50-500Hz frequency bands); S34. Based on the reinforcement learning algorithm of continuous action space, construct a policy network and a value evaluation network, introduce model prediction control force into the reward function, guide the value evaluation or policy update process of the reinforcement learning policy, and train to obtain an active suspension control model.

[0016] Compared with the prior art, the present invention has the following advantages: This invention first constructs a high-frequency quarter-vehicle active suspension model to describe the dynamic relationship between suspension vibration and tire acoustic response. Based on this model, a model predictive control strategy is designed to solve for the model predictive control force that satisfies the system constraints. Then, the model predictive control force is introduced as expert reference information into the reinforcement learning training process to obtain an active suspension control model. Subsequently, the current system state is input into the active suspension control model to output the active suspension control quantity. This invention can achieve synergistic optimization of low-frequency ride comfort and mid-to-high-frequency vibration and in-vehicle noise suppression performance while ensuring system stability and constraint satisfaction. It also alleviates the high-frequency vibration and noise problems caused by the high-frequency components of the control signal in existing electronic suspension control technologies.

[0017] This invention, in constructing a high-frequency quarter-vehicle active suspension model, fully considers the dynamics of the suspension structure, the characteristics of the tire rigid ring, and the modal coupling effect of the tire acoustic cavity. It can accurately describe the dynamic relationship between suspension vibration and tire acoustic response. Compared with traditional suspension models that mainly focus on low-frequency dynamic characteristics, this invention constructs a quarter-vehicle active suspension model that includes the high-frequency structural characteristics of the tire and the vibration-acoustic coupling effect. This expands the dynamic description of the suspension system from the low-frequency range to the mid-to-high frequency vibration and noise band, providing a unified modeling foundation for subsequent full-frequency suspension control.

[0018] This invention utilizes model predictive control force with good mid-to-high frequency control performance as expert knowledge to guide the reinforcement learning process. That is, a predictive control strategy with constraint handling capability is introduced as expert reference information in the reinforcement learning control process. This can effectively suppress unfavorable high-frequency control behavior while maintaining the adaptive advantages of reinforcement learning, and ensure that the overall vibration and noise performance of the control system is optimal across the entire frequency band. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram illustrating the application process of an example. Figure 3 This is a schematic diagram of the vertical acoustic modes in the acoustic cavity of a rotating tire. Figure 4 This is a schematic diagram of the forward and backward acoustic modes in the acoustic cavity of a rotating tire. Figure 5 This is a schematic diagram of a high-frequency quarter-vehicle active suspension model. Figure 6 This is a network architecture diagram of the reinforcement learning agent in the embodiment; Figure 7 This is a framework diagram of the vehicle active suspension control in the embodiment; Figure 8 This is a schematic diagram comparing the simulation test results of the vertical acceleration power spectral density at the wheel center in the embodiment; Figure 9 This is a schematic diagram comparing the simulation test results of the in-vehicle noise frequency domain sound pressure level power spectral density in the embodiment; Figure 10 The above is a frequency domain diagram showing the vertical acceleration of the vehicle body of 1 / 4 of the vehicles in the embodiment under different suspension control strategies on a Class B random road surface. Figure 11 The results show the frequency domain results of the in-vehicle noise sound pressure level of 1 / 4 of the vehicles in the example under different suspension control strategies on Class B random road surfaces. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0021] Example like Figure 1 As shown, a method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension includes the following steps: S1. Construct a high-frequency quarter-vehicle active suspension model to describe the dynamic relationship between suspension vibration and tire acoustic response. S2. Based on the high-frequency quarter-vehicle active suspension model, design a model predictive control strategy and solve for the model predictive control force that satisfies the system constraints. S3. Construct a reinforcement learning controller and introduce the model-predicted control force as expert reference information into the reinforcement learning training process to obtain an active suspension control model. S4. Input the current system state into the active suspension control model and output the active suspension control quantity, which is used to control the working state of the active suspension accordingly.

[0022] This embodiment applies the above-described scheme, and in order to verify the effectiveness of the above-described scheme, this embodiment also establishes a vibration-noise transfer function model to establish a clear vibration-noise correlation between the electronically controlled suspension system and the road noise control system, so as to effectively evaluate the vibration and noise suppression performance of this scheme in the full frequency range.

[0023] like Figure 2 As shown, the application process of this embodiment includes: Step 1: Construct a high-frequency quarter-vehicle active suspension model that includes suspension structure dynamics, tire rigid ring characteristics, and tire acoustic cavity modal coupling effects. This model is used to describe the dynamic relationship between suspension vibration and tire acoustic response and to obtain system state information. Step 2: Identify the dynamic transmission relationship between vehicle body vibration and in-vehicle noise through road tests, and establish a vibration-noise transfer function model; Step 3: Based on the high-frequency vehicle dynamics model constructed in Step 1, design a predictive control strategy to solve for the model predictive control force that satisfies the system constraints; Step 4: Construct a reinforcement learning controller, introduce the model predictive control strategy as expert reference information into the reinforcement learning training process, guide the optimization of the control strategy, and obtain the final active suspension control input; Step 5: Based on the vibration-noise transfer function model established in Step 2, construct structural vibration and noise evaluation indicators related to in-vehicle noise to evaluate the vibration and noise suppression performance of the active suspension control strategy across the entire frequency range.

[0024] The specific process of step one above includes: The characteristics of the tire rigid ring are introduced into the suspension structure dynamics model to describe the structural dynamic behavior of the tire under medium and high frequency operating conditions, thus establishing a high frequency structural dynamics model of the tire-wheel system: in, z s , z u and z b These represent the displacements of the sprung mass, the unsprung mass, and the rigid ring, respectively. m s Represents the sprung mass. m a Represents the unsprung mass of the section without a rigid ring. m b Represents the mass of the rigid ring. k b and c b These represent the wheel-ring stiffness and damping, respectively, connecting the rim and the rigid ring. kct This represents the grounding stiffness connecting the rigid ring and the ground. k s ,and c s These represent suspension stiffness and damping, respectively.

[0025] A tire acoustic cavity model is established, considering the influence of tire structural deformation under static load on the acoustic cavity modes. The dominant acoustic modes that influence the acoustic response during vehicle operation are selected as the modeling objects. The main acoustic modes of the tire acoustic cavity are equivalently modeled as simple harmonic vibration elements, and the original degenerate modes are decomposed into vertical and lateral modes, as shown below. Figure 3 and Figure 4 As shown. In a coordinate system rotating with the tire, the dynamic behavior of the tire acoustic cavity is described by the following equation: in, X and Z This indicates the motion of the center of mass of the enclosed air within the acoustic cavity of a tire; ω 0 represents the natural frequency of the tire's acoustic cavity in the unloaded state; ζ The damping ratio; f 0 represents the static load applied along the Z-axis; a and b Describe the offset of the natural frequencies of the vertical and lateral modes under load (where a < 0, b > 0). c This is a proportionality coefficient. All of the above parameters can be identified through experimental methods. Furthermore, f r This represents the random dynamic component in the tire-road interaction force, which can be represented by... f r = k ct ( z b - q ) was calculated.

[0026] Based on the tire rigid ring model and the tire acoustic cavity model, the mechanical coupling relationship between the acoustic cavity vibration and the tire-wheel system is established, and the equivalent force generated by the acoustic cavity vibration is calculated. F z Introduced into the tire structural dynamics model to characterize the bidirectional coupling effect between the tire acoustic cavity modes and the suspension structure, equivalent forces F z The expression is: in, w e Indicates the effective width of the wheel; rw Indicates the wheel radius; ρ 0 represents the density of the gas inside the tire in equilibrium.

[0027] By integrating the tire structural dynamics model, the tire acoustic cavity model, and the suspension system and vehicle structural dynamics model, a high-frequency quarter-vehicle active suspension model is constructed, incorporating suspension structural dynamics, tire rigid ring characteristics, and the modal coupling effect of the tire acoustic cavity. Figure 5 As shown: In the formula, F a This represents the active suspension control force.

[0028] Obtain vehicle status information, including the vertical acceleration of the sprung mass. Vertical velocity of the sprung mass Suspension dynamic deflection z0 -z u and suspension speed - .

[0029] The specific process of step two above includes: Under full vehicle road testing conditions, vibration sensors and acoustic sensors are arranged along the vibration transmission path of the vehicle structure to simultaneously collect vibration signals at key nodes of the suspension and body. x exp ( t and noise signals at the target location inside the vehicle. y exp ( t This method obtains raw vibration and noise data under multiple operating conditions. In practical applications, the sampling frequency of vibration and noise signals should be no less than twice the upper limit frequency of the frequency band of interest to meet the sampling accuracy requirements for mid-to-high frequency vibration and noise analysis. Key suspension and vehicle body nodes may include, but are not limited to, the wheel center, steering knuckle, lower suspension control arm, upper and lower damper support points, and the connection points between the vehicle body and subframe. Target locations inside the vehicle may include, but are not limited to, locations near the heads of front or rear passengers. The road test conditions are preferably random road surface conditions capable of exciting mid-to-high frequency vibrations and road noise in the vehicle. Road surface types may include, but are not limited to, rough asphalt pavement, small brick pavement, or other typical roads with significant high-frequency excitation characteristics. The test vehicle speed can preferably be set within a speed range where road noise dominates the vehicle's interior noise.

[0030] The collected vibration signals and in-vehicle noise signals are preprocessed and subjected to frequency domain transformation. In this embodiment, the preprocessing includes downsampling and calculation of the first... i The self-power spectral density of a vibration input signalG ii ( f ), and calculate the first i Cross-power spectral density between vibration input signal and in-vehicle noise output signal G iy ( f ). Where * denotes the complex conjugate operation; L Indicates signal length. X i ( f ) indicates the first i The spectrum of an input vibration signal; Y ( f () represents the spectrum of the noise signal inside the vehicle.

[0031] Based on the calculated power spectral density (including self-power spectral density and cross-power spectral density), the frequency domain transfer relationship from suspension vibration input to in-vehicle noise output is estimated using the operating condition transfer path analysis method, and the corresponding vibration-noise frequency domain transfer function model is established as follows: Then, an inverse Fourier transform is performed on the vibration-noise frequency domain transfer function to obtain the equivalent time-domain impulse response of the vibration-noise. h iy ( t This is used to characterize the dynamic mapping relationship between suspension vibration and in-vehicle noise response. The time-domain impulse response and the simulated vehicle vibration signal are compared. x i sim ( t Perform convolution operations to reconstruct the corresponding in-vehicle noise response. ( t ): in, x i sim ( t ) represents the simulated acceleration signal of the vehicle body, and * represents the convolution operation.

[0032] The effectiveness of the vibration-noise transfer function model was verified and corrected by comparing and analyzing the results with those from road tests.

[0033] The specific process of step three above includes: Based on the current system state, a predictive control optimization model containing system constraints is constructed. At each sampling step... k The optimization problem is described as follows: in, N Indicates the length of the prediction time domain. Q Represents the state weighting matrix. R This represents the control input weighting matrix. F This represents the terminal weighting matrix.

[0034] The optimization model is then solved within a finite prediction time domain to obtain the model's predicted control force, which serves as the reference control input for the active suspension. This embodiment solves the optimization model within a finite number of prediction steps to obtain the control sequence. F ( k )={ F ( k | k ), F ( k +1| k ), …, F ( N | k )}, and the first control variable among them. F ( k | k ) as reference control input for active suspension F MPC .

[0035] The specific process of step four above includes: The state observations for constructing a reinforcement learning controller are as follows s t =[ , z 0 -z u , - The state observations include system state information that characterizes the vehicle's low-frequency comfort and mid-to-high-frequency vibration and noise characteristics. This system state information includes at least the vehicle body vibration response. Body abruptness Vehicle speed The relative displacement of the suspension z0 -z u and its changing characteristics - .

[0036] Define the action space of the reinforcement learning controller as the active suspension control force. Fa Its value range is limited by the physical constraints of the suspension structure and actuators. F a ∈[ F min , F max ].

[0037] Construct a composite reward function that includes both time-domain and frequency-domain performance metrics: R ( s t , a t )= R time ( s t , a t )+ R freq ( s t , a t (Among them, time-domain performance indicators) R time ( s t , a t This is used to constrain the low-frequency ride comfort and suspension operating conditions of a vehicle; frequency domain performance indicators. R freq ( s t , a t The two are used to characterize the vibration response characteristics related to in-vehicle noise within the target frequency band, and their expressions are as follows: In the formula, W 1. W 2. W 3. W 4. W 5 represents the time-domain reward weighting coefficient, which in this embodiment is taken as 2, 1, 0.5, 0.5, and 0.1 respectively. A norm , L norm , J norm , F norm The normalization coefficients are 4, 0.01, 1500, and 500 in this embodiment. Let be the vertical acceleration of the sprung mass. z s – za For suspension dynamic deflection, The vertical jerk of the sprung mass. W 0-20 , W 20-50 , W 50-500 The frequency domain reward weighting coefficients are 2.5, 2.5, and 5 respectively in this embodiment. RMS 0-20Hz , RMS 20-50Hz and RMS 50-500Hz These represent the root mean square values ​​of the sprung mass acceleration in the frequency bands of 0-20 Hz, 20-50 Hz, and 50-500 Hz, respectively.

[0038] A reinforcement learning algorithm based on continuous action space is used to construct a policy network and a value evaluation network, with the network structure as follows: Figure 6 As shown in the figure, T This represents the window length for constructing the timing input signal, with 100 sampling points. C in The number of channels representing the input state quantity is consistent with the number of observations; in this embodiment, it is set to 5. Both the policy network and the value evaluation network include a temporal feature extraction module based on Long Short Term Memory (LSTM) networks, used to extract temporally relevant features from suspension state observations at consecutive time steps. The temporal feature extraction module in the policy network employs a two-layer stacked LSTM structure to process the input multi-channel state time series and selects the hidden state at the last time step as the feature representation. The feature representation is flattened and input to three fully connected layers. Finally, the output layer generates continuous control actions, and the output is normalized using a nonlinear activation function and scaled to meet action space constraints.

[0039] The value assessment network employs the same LSTM-based temporal feature extraction structure as the policy network to process the system state inputs and, in conjunction with control action information, estimate the long-term reward of the current state-action pair. The final system control block diagram is shown below. Figure 7 As shown, expert information from the predictive control policy output is incorporated into the reward function. F MPC This guides the value assessment or strategy updating process of reinforcement learning strategies.

[0040] Using the reinforcement learning control strategy obtained from training, active suspension control input is generated online based on the current system state. Under the premise of satisfying system stability and constraints, this control input is used to achieve coordinated suppression of vehicle vibration and in-vehicle noise.

[0041] The specific process of step five above includes: Based on the vehicle vibration response signal obtained under active suspension control, key state variables that can reflect vehicle ride comfort and safety are selected to evaluate the vibration control effect of the active suspension.

[0042] By using the vibration-noise transfer function model, the vehicle body vibration response signal is mapped to the corresponding in-vehicle noise response signal to characterize the degree of influence of active suspension vibration control on in-vehicle noise.

[0043] The vibration response signal and the corresponding in-vehicle noise response signal are divided and analyzed by frequency band. The suspension vibration response is analyzed in the low frequency band of 0-20Hz and the in-vehicle noise response is analyzed in the mid-high frequency band of 20-500Hz, so as to quantitatively characterize the effect of active suspension control strategy on vibration and noise suppression in different frequency bands.

[0044] In this embodiment, the vehicle parameters are set as follows: sprung mass m s = 388 kg, unsprung mass of the section without rigid rings m a = 29 kg, mass of rigid ring m b = 5 kg, wheel-ring stiffness connecting the rim and the rigid ring k b = 1800000 N / m, wheel-ring damping connecting the rim and the rigid ring. c b = 1000 N·s / m, grounding stiffness connecting the rigid ring and the ground. k ct = 100,000 N / m, suspension stiffness k s = 20000, suspension damping c s = 1500, u min =-1000 N, u max =1000 N. The tire acoustic cavity parameters are set as follows: natural frequency of the unloaded tire acoustic cavity. ω 0 = 210 × 2π rad / s, tire dynamic load coefficient c =90, effective wheel width w e=0.205m, wheel radius r w =0.3184m, the gas density inside the tire under equilibrium conditions. ρ 0 = 3.76 kg / m 3 Damping ratio ζ= 0.001, a = -0.6, b = 0.01. During simulation and control implementation, the control force output update frequency of the active suspension controller is set to 100 Hz to reflect the low control frequency caused by delays in actual electronically controlled suspensions. The numerical simulation frequency of the vehicle dynamics model is set to 1000 Hz. Based on these parameter settings, the established model can simultaneously characterize the vehicle body vibration characteristics dominated by the suspension system in the low-frequency range, and the resonance peak characteristics formed by the combined effects of unsprung mass, rigid ring tire structure, tire acoustic cavity modes, and the main frequency of the control force in the mid-to-high frequency range.

[0045] In this embodiment, the road test condition used to identify the vibration-noise transfer function is set as follows: the vehicle travels at a constant speed of 60 km / h on a rough asphalt road surface. The vibration signal measurement location is selected at the center of the right front wheel, and the in-vehicle noise measurement location is selected approximately 10 cm in front of the passenger seat headrest. During the road test, the sampling frequency of the vibration signal is set to 51200 Hz; the corresponding sampling frequency of the simulation model output signal is 10000 Hz. To facilitate a consistent comparison between the simulation results and the experimental results, both the experimental and simulation signals are resampled to 1000 Hz before frequency domain analysis. Figure 8 , Figure 9 As shown, the simulation model can effectively capture the main frequency characteristics of the noise transmitted from tire vibration to the vehicle interior, and the key peak frequency is consistent with the experimental results.

[0046] In this embodiment, the simulation conditions used to verify the effectiveness of the control strategy are set as follows: the vehicle speed is 60km / h, the road excitation is a Class B random road surface, and the vibration and noise suppression effect of the whole frequency band is simulated and verified. The proposed method is compared with the simplified 1 / 4 car model passive suspension, the high-frequency 1 / 4 car passive suspension (passive-ACO), the model predictive control (MPC), and the single reinforcement learning control strategy (Pure-RL).

[0047] like Figure 10 As shown, the reinforcement learning-model predictive control hybrid control strategy can significantly reduce the power spectral density of the vehicle's vertical acceleration across the entire frequency range. This control strategy exhibits good vibration suppression performance in the mid-to-high frequency range. Figure 10As shown in section (a.1), the main reason is that the equivalent damping level of the unsprung mass mode is improved by the control action, which significantly reduces the relevant resonance peak.

[0048] like Figure 11 As shown, in the passive suspension response considering the tire acoustic cavity mode, a significant resonance peak appears around 230 Hz. This resonance peak is transmitted into the vehicle interior through the structural transmission path, resulting in a significant increase in sound pressure level above 200 Hz. Compared to the response results of different control strategies, the hybrid control strategy proposed in this invention exhibits a superior overall effect in suppressing in-vehicle noise, effectively attenuating vibration and noise across multiple frequency bands.

[0049] In summary, this embodiment effectively characterizes the in-vehicle noise response under suspension control in a simulation environment by introducing the characteristics of the tire rigid ring and the modal coupling effect of the tire acoustic cavity into the quarter-vehicle suspension model and combining it with the vibration-noise transfer function identified from road tests. Furthermore, by utilizing a model predictive control strategy with good mid-to-high frequency control performance as an expert knowledge-guided reinforcement learning process, it can achieve synergistic optimization of low-frequency ride comfort and mid-to-high frequency vibration and in-vehicle noise suppression performance while ensuring system stability and constraint satisfaction. This provides an effective technical approach for the integrated design of electronically controlled suspension and active road noise control, and has promising engineering application prospects.

[0050] This study constructs a quarter-vehicle model that incorporates the high-frequency structural characteristics of tires and the vibration-audio coupling effect, extending the dynamic description of the suspension system from the low-frequency range to the mid-to-high frequency vibration and noise bands, providing a unified modeling foundation for full-band suspension control. In the control strategy design, a predictive control strategy with constraint handling capabilities is introduced as expert reference information to guide the reinforcement learning control process. This effectively suppresses unfavorable high-frequency control behaviors while maintaining the adaptive advantages of reinforcement learning, ensuring optimal overall vibration and noise performance of the control system across the entire frequency band. Furthermore, a clear vibration-audio correlation is established between the electronically controlled suspension system and the road noise control system, facilitating the coordinated optimization design of suspension control parameters, control strategies, and overall vehicle vibration and noise performance. This approach balances low-frequency comfort with mid-to-high frequency noise suppression performance, effectively mitigating the high-frequency vibration and noise problems caused by the high-frequency components of the control signal in existing electronic suspension control technologies.

Claims

1. A method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension, characterized in that, Includes the following steps: S1. Construct a high-frequency quarter-vehicle active suspension model to describe the dynamic relationship between suspension vibration and tire acoustic response. S2. Based on the high-frequency quarter-vehicle active suspension model, design a model predictive control strategy and solve for the model predictive control force that satisfies the system constraints. S3. Construct a reinforcement learning controller and introduce the model-predicted control force as expert reference information into the reinforcement learning training process to obtain an active suspension control model. S4. Input the current system state into the active suspension control model and output the active suspension control quantity, which is used to control the working state of the active suspension accordingly.

2. The method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension according to claim 1, characterized in that, Specifically, S1 is a high-frequency quarter-vehicle active suspension model constructed based on suspension structure dynamics, tire rigid ring characteristics, and tire acoustic cavity modal coupling effect.

3. The method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension according to claim 2, characterized in that, S1 includes the following process: S11. Introduce the tire rigid ring characteristics into the suspension structure dynamics model to describe the structural dynamic behavior of the tire under medium and high frequency operating conditions, and establish a high frequency structural dynamics model of the tire-wheel system. S12. Considering the influence of the structural deformation of the tire under static load on the acoustic cavity mode, the main acoustic modes of the tire acoustic cavity are equivalently modeled as simple harmonic dynamic elements, and the original degenerate mode is decomposed into vertical mode and lateral mode. The acoustic mode that plays a dominant role in the acoustic response under vehicle operation is selected as the modeling object. The dynamic behavior of the tire acoustic cavity is described in the coordinate system that rotates with the tire to establish the tire acoustic cavity model. S13. Based on the tire rigid ring model and the tire acoustic cavity model, establish the mechanical coupling relationship between the acoustic cavity vibration and the tire-wheel system, and introduce the equivalent force generated by the acoustic cavity vibration into the tire structure dynamics model to characterize the bidirectional coupling effect between the tire acoustic cavity mode and the suspension structure. S14. Integrate the tire structure dynamics model, tire acoustic cavity model, suspension system and vehicle body structure dynamics model to construct a high-frequency quarter vehicle active suspension model.

4. The method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension according to claim 3, characterized in that, The high-frequency structural dynamics model of the tire-wheel system in S11 is as follows: in, z s , z u and z b These represent the displacements of the sprung mass, the unsprung mass, and the rigid ring, respectively. m s Represents the sprung mass. m a Represents the unsprung mass of the section without a rigid ring. m b Represents the mass of the rigid ring. k b and c b These represent the wheel-ring stiffness and damping, respectively, connecting the rim and the rigid ring. k ct This represents the grounding stiffness connecting the rigid ring and the ground. k s ,and c s These represent suspension stiffness and damping, respectively.

5. The method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension according to claim 4, characterized in that, The tire acoustic cavity model in S12 is specifically as follows: in, X and Z This indicates the motion of the center of mass of the enclosed air within the acoustic cavity of a tire; ω 0 represents the natural frequency of the tire's acoustic cavity in the unloaded state; ζ The damping ratio; f 0 represents the static load applied along the Z-axis; a and b Describe the offset of the natural frequencies of the vertical and lateral modes under load, respectively, where a < 0, b > 0; c This is the proportionality coefficient. f r This represents the random dynamic component in the tire-road interaction force.

6. The method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension according to claim 5, characterized in that, The equivalent force generated by the vibration of the acoustic cavity in S13 is specifically as follows: in, w e Indicates the effective width of the wheel; r w Indicates the radius of the wheel; ρ 0 represents the density of the gas inside the tire in equilibrium.

7. The method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension according to claim 6, characterized in that, The high-frequency quarter-vehicle active suspension model in S14 is specifically as follows: in, F a This is for active suspension control.

8. The method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension according to claim 1, characterized in that, S2 includes the following process: S21. Based on the high-frequency quarter-vehicle active suspension model, obtain system state information to construct a predictive control optimization model containing system constraints. S22. Solve the predictive control optimization model within the finite prediction time domain to obtain the model predictive control force, which is used as the reference control input for the active suspension.

9. A method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension according to claim 8, characterized in that, The system status information includes vehicle acceleration, suspension travel, and tire dynamic load, specifically vehicle vibration response. Body abruptness Vehicle speed The relative displacement of the suspension z0 -z u and its changing characteristics - .

10. A method for suppressing vibration and noise across the entire frequency band of a vehicle's active suspension according to claim 9, characterized in that, S3 includes the following process: S31. Construct the state observations of the reinforcement learning controller as follows: s t =[ , z 0 -z u , - The state observations include system state information that characterizes the low-frequency comfort and mid-to-high-frequency vibration and noise characteristics of the vehicle. S32. Define the action space of the reinforcement learning controller as the active suspension control force. F a Its value range is limited by the physical constraints of the suspension structure and actuators. F a ∈[ F min , F max ]; S33. Construct a composite reward function that includes time-domain performance metrics and frequency-domain performance metrics. R ( s t , a t )= R time ( s t , a t )+ R freq ( s t , a t (Among them, time-domain performance indicators) R time ( s t , a t This is used to constrain the low-frequency ride comfort and suspension operating conditions of a vehicle; frequency domain performance indicators. R freq ( s t , a t (This is used to characterize the vibration response characteristics related to in-vehicle noise within a target frequency band, which includes the 0-20Hz, 20-50Hz and 50-500Hz frequency bands); S34. Based on the reinforcement learning algorithm of continuous action space, construct a policy network and a value evaluation network, introduce model prediction control force into the reward function, guide the value evaluation or policy update process of the reinforcement learning policy, and train to obtain an active suspension control model.

Images